CN109101945A - A kind of more detection method of license plate in traffic video monitoring image - Google Patents
A kind of more detection method of license plate in traffic video monitoring image Download PDFInfo
- Publication number
- CN109101945A CN109101945A CN201810980740.4A CN201810980740A CN109101945A CN 109101945 A CN109101945 A CN 109101945A CN 201810980740 A CN201810980740 A CN 201810980740A CN 109101945 A CN109101945 A CN 109101945A
- Authority
- CN
- China
- Prior art keywords
- component
- filter
- value
- image
- channel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/625—License plates
Abstract
The invention discloses more detection method of license plate in a kind of traffic video monitoring image, belong to the technical field of image procossing, mainly comprise the steps that (1) pre-processes the original video data of Traffic Surveillance Video data, to obtain number plate of vehicle target object；(2) using deformable part model training number plate of vehicle target object；(3) the application training aspect of model and number plate of vehicle target object carry out characteristic matching, to complete to detect, with reach can to license plate recognition accuracy and efficiency in terms of be obviously improved, simplify the purpose of operation complexity.
Description
Technical field
The invention belongs to the technical fields of image procossing, in particular to more in a kind of traffic video monitoring image
Detection method of license plate.
Background technique
Traffic video monitoring based on Fast Monitoring, monitoring point is distributed in wagon flow, the stream of people compare concentration intersection,
Road traffic situation is uploaded to road monitoring command centre, center watch by image transmitting channel by emphasis section in real time
Member can understand each region pavement behavior in time accordingly, to adjust each crossing vehicle flow, it is ensured that traffic is unobstructed.To monitoring road
The situation violating the regulations of face vehicle, can find in time and arrange processing road traffic accident etc., and can be all kinds of for traffic, public security etc.
The detection of case provides technical support, greatly improves the level and efficiency of public security organ's law enforcement.
The intellectual technology of Tsinghua University and system National Key Laboratory have been devoted to the research of License Plate, more early
Simple and direct practical high effective model search is designed using BOES (Bionic Object Exploring Strategy) strategy to calculate
License plate is accurately positioned in method, and then, which, which proposes, combines mathematical morphology and character stroke point under a kind of multiresolution
The algorithm of locating license plate of vehicle of analysis, and indicate that the algorithm is spaced license plate area contrast is lower between the every character of license plate number
Locating effect undesirable deficiency when larger, but the business that the detection accuracy of this method is not met by traffic video monitoring needs
It asks.
Central China University of Science and Technology's Control Science and Engineering lie in propose within 2000 it is a kind of based on vertical characters boundary characteristic
License plate locating method, give improve License Plate precision and generalization ability measure, but the computational complexity of its algorithm compared with
Height is unable to satisfy the real-time traffic demands of field of traffic.
To sum up, there is answering to the accuracy rate of Car license recognition, efficiency and operation for traffic video monitoring image at present
There is biggish deficiency on miscellaneous degree, is urgent problem to be solved in traffic video monitoring.
Summary of the invention
In consideration of it, in order to solve the above problems existing in the present technology, the present invention provides a kind of traffic video monitoring figures
As in more detection method of license plate with reach can to license plate recognition accuracy and efficiency in terms of be obviously improved, simplify
The purpose of operation complexity.
The technical scheme adopted by the invention is as follows: more detection method of license plate in a kind of traffic video monitoring image, mainly
The following steps are included:
(1) original video data of Traffic Surveillance Video data is pre-processed, to obtain number plate of vehicle target object；
(2) using deformable part model training number plate of vehicle target object；
(3) the application training aspect of model and number plate of vehicle target object carry out characteristic matching, to complete to detect.
Further, specific step is as follows for the step (1):
1) histogram equalization processing is carried out to original video data, to show number plate of vehicle target pair to be detected
As；
2) denoising is carried out by median filtering method to the video image that step 1) obtains.
Further, detailed process is as follows for the step 1):
It is F to the i-th frame imagei, with FiIn R value, G value, the B value of all pixels point carry out statistics classification, to the channel R, G
Three channel, channel B channel images carry out histogram equalization；Then, each channel is merged, it is colored to constitute a width
Image.
Further, detailed process is as follows for the step 2):
A rectangular filter template is set, is moved using video image of the rectangular filter template to step 1), with Fi
(x, y, r) or Fi(x, y, g) or Fi(x, y, b) is the intermediate value of rectangular filter template position surrounding pixel values and should
It is worth the pixel value as rectangular filter template center point pixel；Wherein, Fi(x, y, r) or Fi(x, y, g) or Fi(x,y,b)
Respectively represent image FiThe pixel value in the channel R, the channel G, channel B at two-dimensional position (x, y).
Further, specific step is as follows for the step (2):
A, it is based on histograms of oriented gradients HOG algorithm, extracts the i-th frame image FiHOG feature pyramid, enable HOG feature
Pyramid is global feature；
B, using the feature of SVM model training number plate of vehicle target object, number plate of vehicle target object is set as by m
The model that component is constituted, the component of each component model are n；Positive sample is enabled to integrate as PS, negative sample integrates as NS, number plate of vehicle mesh
The model training process for marking object is as follows:
(a), the callout box in positive sample collection PS is roughly divided into m class according to shooting angle；
(b), with the root filter P of m component model of SVM model training1, P2... .., Pm；
(c), it is polymerize based on image space, m component model of training in step (b) is transformed into one without containing component
Initial root filter, meanwhile, by the root filter of m component model respectively by being interpolated into the initialization of twice resolution space
Obtain n component filter；
(d), the initial root filter that goes matching step (c) to obtain respectively with positive sample collection and negative sample collection simultaneously calculates root filter
Wave device is scored at R (x, y)；For the position of positive sample image training highest scoring, replaced with newest location parameter old
Location parameter obtains new positive sample image with this；To negative sample image, old sample is replaced with the minimum sample of training score
Image, to obtain new negative sample image；
(e), step (d) is repeated, when the energy maximum of component filter and the intersecting area of root filter, corresponding position
The final position that parameter is component is set, all training obtains component feature by the position of n component according to this process.
Further, detailed process is as follows for the step (3):
By global feature and component feature respectively with root filter and component filter convolution, then to component filter
Response is sampled, and makes it to be weighted and averaged under same scale with the response of root filter, is carried out according to formula (1) comprehensive
The calculating of score is closed, the position of highest scoring is the horizontal axis of target, ordinate of orthogonal axes position:
Wherein, R (x, y) is the score of root filter, Di,lWhen being placed on l layers of position (x, y) for the anchor point of i-th of component,
Its maximum contribution value to root position score, l indicate the pyramidal number of plies number of HOG feature.
Further, it is filtered in the step (e) by using the zone of action of energy function judgement part filter and root
The Energy maximum value of the zone of action intersecting area of wave device, energy function are as follows:
P=R2+G2+B2
Wherein, P represents energy value, and its initial value is 0；R, G, B respectively represent the i-th frame image FiInstitute in middle intersecting area
There are the R value, G value, B value of pixel.
The invention has the benefit that
1, present invention employs deformable part model training number plate of vehicle target objects, in the accuracy rate and efficiency of identification
Upper to have greater advantage, which has obtained multiple authentication in target detection.
2, using the model of hub-and-spoke configuration, model consists of three parts deformable part model in the present invention: one
Relative coarseness but the root filter that entire target can be covered, the higher component filter of one group of resolution ratio and component filter phase
For the spatial position of root filter.Wherein, root filter can describe the global feature of target, and component filter then can
The more fine feature in target part is described, resolution ratio is 2 times of root filter.Meanwhile the model is adopted in the detection process
With Pyramidal search mode, therefore, the computational complexity of the model can satisfy calculates demand in real time.Meanwhile the mould
Type can be run again by single, and detection while realizing multiple number plate of vehicle targets saves detection time.
Detailed description of the invention
Fig. 1 is number plate of vehicle target object in more detection method of license plate in traffic video monitoring image provided by the invention
Model training flow chart.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, practical below in conjunction with this is newly to make
The purposes, technical schemes and advantages of the embodiment of the present invention are clearer, technical solution in the embodiment of the present invention carry out it is clear,
It is fully described by, it is clear that described embodiments are some of the embodiments of the present invention, instead of all the embodiments.
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the present invention can phase
Mutually combination.
Definition and english abbreviation to symbolic variable in the present embodiment illustrate as follows:
1, original video data: VD；
2, a certain frame in video stream data (the i-th frame) image: Fi；
3, the channel R: red color tone channel；The channel G: green hue channel；Channel B: blue cast channel；
4, (x, y) indicates a certain frame (the i-th frame) image FiTwo-dimentional horizontal axis, ordinate of orthogonal axes；
5、Fi(x, y) indicates image FiPixel value vector at two-dimensional position (x, y)；
6、Fi(x,y,r)、Fi(x,y,g)、Fi(x, y, b) respectively represents image FiThe channel R at two-dimensional position (x, y),
The channel G, channel B pixel value；
7, HOG algorithm (Histogram of Oriented Gradient, HOG): histograms of oriented gradients；
8, SVM model (Support Vector Machine, SVM) --- support vector machines.
The present invention provides more detection method of license plate in a kind of traffic video monitoring image, mainly comprise the steps that
(1) original video data of Traffic Surveillance Video data being pre-processed, original video data is indicated with VD,
To obtain number plate of vehicle target object, specific step is as follows for the step:
1) histogram equalization processing is carried out to original video data, to show number plate of vehicle target pair to be detected
As；Detailed process is as follows for the step:
It is F to the i-th frame imagei, with FiIn R value, G value, the B value of all pixels point carry out statistics classification, to the channel R, G
Three channel, channel B channel images carry out histogram equalization；Then, each channel is merged, it is colored to constitute a width
Image.
2) video image (color image constituted in step 1)) that step 1) obtains is carried out by median filtering method
Denoising, detailed process is as follows for the step:
A rectangular filter template is set, template size is 5*5 in the present embodiment, using rectangular filter template to step
1) obtained video image is moved, with Fi (x, y, r) or Fi (x, y, g) or Fi (x, y, b) for rectangular filter mould
The intermediate value of plate position surrounding pixel values and using the value as the pixel value of rectangular filter template center point pixel；Wherein, Fi
(x, y, r) or Fi (x, y, g) or Fi (x, y, b) respectively represent image FiThe channel R, G at two-dimensional position (x, y) is logical
The pixel value in road, channel B.
(2) using deformable part model training number plate of vehicle target object；Specific step is as follows for the step:
A, it is based on histograms of oriented gradients HOG algorithm, integrates two kinds of gradient formers: having symbol gradient and without symbol gradient,
The feature vector of a 3* (9+18)=81 dimension is obtained, by carrying out principal component analysis and parsing dimensionality reduction to it, by original 81
Dimensional feature add up to each row and column respectively is down to 31 dimensions, extracts the i-th frame image FiHOG feature pyramid, enable HOG feature
Pyramid is global feature；
Deformable part model will carry out object detection and identification based on sliding window, using HOG character description method, and
Establish HOG feature pyramid comprising: establish image pyramid and every layer of generation HOG feature.
B, as shown in the flowchart of fig.1, using the feature of SVM model training number plate of vehicle target object, by number plate of vehicle
Target object is set as the model being made of m component, takes m=3 in the present embodiment, and the component of each component model is n,
N=5 is taken in the present embodiment；Positive sample is enabled to integrate as PS, negative sample integrates as NS, wherein positive sample collection is to include target object
Image pattern, negative sample collection are the image pattern not comprising target object.
The model training process of number plate of vehicle target object is as follows:
(a), the callout box in positive sample collection PS is roughly divided into 3 classes according to shooting angle, in the present embodiment, respectively
Basic horizontal, be tilted to the left, be tilted to the right three classes；
(b), with the root filter P of 3 component models of SVM model training1, P2, P3；
(c), it is polymerize based on image space, 3 component models of training in step (b) is transformed into one without containing component
Initial root filter, meanwhile, the root filter of 3 component models is carried out just by being interpolated into twice resolution space respectively
Beginningization, each component model obtain 5 component filters.Wherein, the zone of action of root filter is entire detection window, substantially
Cover entire number plate of vehicle target object；Component filter be placed in HOG feature it is pyramidal under it is several layers of, cover number plate of vehicle target
Smaller component in object has higher resolution ratio；
(d), the initial root filter that goes matching step (c) to obtain respectively with positive sample collection and negative sample collection simultaneously calculates root filter
Wave device is scored at R (x, y), is defined as: filter vector with the feature vector for the child window that (x, y) is upper left angle point
Dot product.Filter is to be passed through filtering with the rectangular filter template for being used to formulate weight of a child window in HOG pyramid and determined
Whether interested mode is contained in child window, and filter vector is the weight vectors of filter；For the training of positive sample image
The position of highest scoring is replaced old location parameter with newest location parameter (x, y), new positive sample image is obtained with this；
To negative sample image, replace old sample image with the minimum sample of training score, equally, also with newest location parameter (x,
Y) old location parameter is replaced, to obtain new negative sample image；
(e), step (d) is repeated, when the energy maximum of component filter and the intersecting area of root filter, intersecting area
The intersecting area of the zone of action of the zone of action and root filter that are defined as component filter between the two, corresponding position ginseng
Number is the final position of component, and according to this process, by the position of n component, all training obtains component feature.
Wherein, by using the zone of action of energy function judgement part filter and root filter in the step (e)
Zone of action intersecting area Energy maximum value, energy function is as follows:
P=R2+G2+B2
Wherein, P represents energy value, and its initial value is 0；R, G, B respectively represent the i-th frame image FiInstitute in middle intersecting area
There are the R value, G value, B value of pixel.
(3) the application training aspect of model and number plate of vehicle target object carry out characteristic matching, described to be somebody's turn to do to complete to detect
Detailed process is as follows for step:
By global feature and component feature respectively with root filter and component filter convolution, then to component filter
Response is sampled, and makes it to be weighted and averaged under same scale with the response of root filter, is carried out according to formula (1) comprehensive
The calculating of score is closed, the position of highest scoring is the horizontal axis of target, ordinate of orthogonal axes position:
Wherein, R (x, y) is the score of root filter, Di,lWhen being placed on l layers of position (x, y) for the anchor point of i-th of component,
Its maximum contribution value to root position score, l indicate the pyramidal number of plies number of HOG feature.
The present invention is not limited to above-mentioned optional embodiment, anyone can show that other are various under the inspiration of the present invention
The product of form, however, make any variation in its shape or structure, it is all to fall into the claims in the present invention confining spectrum
Technical solution, be within the scope of the present invention.
Claims (7)
1. more detection method of license plate in a kind of traffic video monitoring image, which is characterized in that mainly comprise the steps that
(1) original video data of Traffic Surveillance Video data is pre-processed, to obtain number plate of vehicle target object；
(2) using deformable part model training number plate of vehicle target object；
(3) the application training aspect of model and number plate of vehicle target object carry out characteristic matching, to complete to detect.
2. more detection method of license plate in traffic video monitoring image according to claim 1, which is characterized in that the step
Suddenly (1) specific step is as follows:
1) histogram equalization processing is carried out to original video data, to show number plate of vehicle target object to be detected；
2) denoising is carried out by median filtering method to the video image that step 1) obtains.
3. more detection method of license plate in traffic video monitoring image according to claim 2, which is characterized in that the step
It is rapid that 1) detailed process is as follows:
It is F to the i-th frame imagei, with FiIn R value, G value, the B value of all pixels point carry out statistics classification, to the channel R, the channel G,
Three channel images of channel B carry out histogram equalization；Then, each channel is merged, to constitute a width color image.
4. more detection method of license plate in traffic video monitoring image according to claim 3, which is characterized in that the step
It is rapid that 2) detailed process is as follows:
A rectangular filter template is set, is moved using video image of the rectangular filter template to step 1), with Fi(x,y,
Or F r)i(x, y, g) or Fi(x, y, b) be rectangular filter template position surrounding pixel values intermediate value and using the value as
The pixel value of rectangular filter template center point pixel；Wherein, Fi(x, y, r) or Fi(x, y, g) or Fi(x, y, b) generation respectively
Table image FiThe pixel value in the channel R, the channel G, channel B at two-dimensional position (x, y).
5. more detection method of license plate in traffic video monitoring image according to claim 1, which is characterized in that the step
Suddenly (2) specific step is as follows:
A, it is based on histograms of oriented gradients HOG algorithm, extracts the i-th frame image FiHOG feature pyramid, enable HOG feature pyramid
For global feature；
B, using the feature of SVM model training number plate of vehicle target object, number plate of vehicle target object is set as by m component
The model of composition, the component of each component model are n；Positive sample is enabled to integrate as PS, negative sample integrates as NS, number plate of vehicle target pair
The model training process of elephant is as follows:
(a), the callout box in positive sample collection PS is roughly divided into m class according to shooting angle；
(b), with the root filter P of m component model of SVM model training1, P2... .., Pm；
(c), it is polymerize based on image space, m component model of training in step (b) is transformed into one and does not contain the first of component
Beginning root filter, meanwhile, the root filter of m component model is obtained by being interpolated into twice of resolution space initialization respectively
N component filter；
(d), the initial root filter that goes matching step (c) to obtain respectively with positive sample collection and negative sample collection simultaneously calculates root filter
It is scored at R (x, y)；For the position of positive sample image training highest scoring, old position is replaced with newest location parameter
Parameter obtains new positive sample image with this；To negative sample image, old sample graph is replaced with the minimum sample of training score
Picture, to obtain new negative sample image；
(e), step (d) is repeated, when the energy maximum of component filter and the intersecting area of root filter, corresponding position ginseng
Number is the final position of component, all trains the position of n component to obtain component feature according to this process.
6. more detection method of license plate in traffic video monitoring image according to claim 4, which is characterized in that the institute
Stating step (3), detailed process is as follows:
By global feature and component feature respectively with root filter and component filter convolution, then to the response of component filter
It is sampled, makes it to be weighted and averaged under same scale with the response of root filter, carried out according to formula (1) comprehensive
The calculating divided, the position of highest scoring is the horizontal axis of target, ordinate of orthogonal axes position:
Wherein, R (x, y) is the score of root filter, Di,lWhen being placed on l layers of position (x, y) for the anchor point of i-th of component, it is right
The maximum contribution value of root position score, l indicate the pyramidal number of plies number of HOG feature.
7. more detection method of license plate in traffic video monitoring image according to claim 4, which is characterized in that the step
Suddenly by using the zone of action intersecting area of the zone of action and root filter of energy function judgement part filter in (e)
Energy maximum value, energy function are as follows:
P=R2+G2+B2
Wherein, P represents energy value, and its initial value is 0；R, G, B respectively represent the i-th frame image FiAll pictures in middle intersecting area
R value, G value, the B value of vegetarian refreshments.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810980740.4A CN109101945A (en) | 2018-08-27 | 2018-08-27 | A kind of more detection method of license plate in traffic video monitoring image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810980740.4A CN109101945A (en) | 2018-08-27 | 2018-08-27 | A kind of more detection method of license plate in traffic video monitoring image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109101945A true CN109101945A (en) | 2018-12-28 |
Family
ID=64851273
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810980740.4A Pending CN109101945A (en) | 2018-08-27 | 2018-08-27 | A kind of more detection method of license plate in traffic video monitoring image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109101945A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110636209A (en) * | 2019-04-28 | 2019-12-31 | 刘晓英 | Auxiliary signal parameter analysis method |
| CN111382736A (en) * | 2018-12-29 | 2020-07-07 | 浙江宇视科技有限公司 | License plate image acquisition method and device |
| CN111460996A (en) * | 2020-03-31 | 2020-07-28 | 东南大学 | Truck number plate identification method based on extreme learning deep network fusion model |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104134071A (en) * | 2014-06-20 | 2014-11-05 | 北京工业大学 | Deformable part model object detection method based on color description |
| CN104715239A (en) * | 2015-03-12 | 2015-06-17 | 哈尔滨工程大学 | Vehicle color identification method based on defogging processing and weight blocking |
| CN104751190A (en) * | 2015-04-23 | 2015-07-01 | 武汉大学 | Vehicle part positioning method for vehicle fine recognition |
| US20150363670A1 (en) * | 2014-06-12 | 2015-12-17 | Panasonic Intellectual Property Management Co., Ltd. | Image recognition method and camera system |
| CN105469091A (en) * | 2015-12-03 | 2016-04-06 | 深圳市捷顺科技实业股份有限公司 | Vehicle recognition method and device |
| CN107944359A (en) * | 2017-11-14 | 2018-04-20 | 中电数通科技有限公司 | Flame detecting method based on video |
-
2018
- 2018-08-27 CN CN201810980740.4A patent/CN109101945A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150363670A1 (en) * | 2014-06-12 | 2015-12-17 | Panasonic Intellectual Property Management Co., Ltd. | Image recognition method and camera system |
| CN104134071A (en) * | 2014-06-20 | 2014-11-05 | 北京工业大学 | Deformable part model object detection method based on color description |
| CN104715239A (en) * | 2015-03-12 | 2015-06-17 | 哈尔滨工程大学 | Vehicle color identification method based on defogging processing and weight blocking |
| CN104751190A (en) * | 2015-04-23 | 2015-07-01 | 武汉大学 | Vehicle part positioning method for vehicle fine recognition |
| CN105469091A (en) * | 2015-12-03 | 2016-04-06 | 深圳市捷顺科技实业股份有限公司 | Vehicle recognition method and device |
| CN107944359A (en) * | 2017-11-14 | 2018-04-20 | 中电数通科技有限公司 | Flame detecting method based on video |
Non-Patent Citations (2)
| Title |
|---|
| 钱月晶: "基于可变形部件模型的车牌检测", 《浙江工贸职业技术学院学报》 * |
| 陈芝垚: "基于DPM的行人检测和行人特征提取算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111382736A (en) * | 2018-12-29 | 2020-07-07 | 浙江宇视科技有限公司 | License plate image acquisition method and device |
| CN111382736B (en) * | 2018-12-29 | 2023-05-23 | 浙江宇视科技有限公司 | License plate image acquisition method and device |
| CN110636209A (en) * | 2019-04-28 | 2019-12-31 | 刘晓英 | Auxiliary signal parameter analysis method |
| CN110636209B (en) * | 2019-04-28 | 2020-07-14 | 侯小芳 | Auxiliary signal parameter analysis method |
| CN111460996A (en) * | 2020-03-31 | 2020-07-28 | 东南大学 | Truck number plate identification method based on extreme learning deep network fusion model |
| CN111460996B (en) * | 2020-03-31 | 2022-02-15 | 东南大学 | Truck number plate identification method based on extreme learning deep network fusion model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wei et al. | Multi-vehicle detection algorithm through combining Harr and HOG features | |
| CN104378582B (en) | A kind of intelligent video analysis system and method cruised based on Pan/Tilt/Zoom camera | |
| CN103605977B (en) | Extracting method of lane line and device thereof | |
| CN108549926A (en) | A kind of deep neural network and training method for refining identification vehicle attribute | |
| CN105005989B (en) | A kind of vehicle target dividing method under weak contrast | |
| CN105354568A (en) | Convolutional neural network based vehicle logo identification method | |
| CN104809461B (en) | The licence plate recognition method and system of binding sequence image super-resolution rebuilding | |
| CN109101945A (en) | A kind of more detection method of license plate in traffic video monitoring image | |
| CN111915583B (en) | Vehicle and pedestrian detection method based on vehicle-mounted thermal infrared imager in complex scene | |
| CN106778687A (en) | Method for viewing points detecting based on local evaluation and global optimization | |
| CN103247059A (en) | Remote sensing image region of interest detection method based on integer wavelets and visual features | |
| CN103605953A (en) | Vehicle interest target detection method based on sliding window search | |
| CN111582339B (en) | Vehicle detection and recognition method based on deep learning | |
| CN105184779A (en) | Rapid-feature-pyramid-based multi-dimensioned tracking method of vehicle | |
| CN107704853A (en) | A kind of recognition methods of the traffic lights based on multi-categorizer | |
| CN103530608A (en) | Vehicle type judgment method and vehicle type judgment device | |
| CN107895139A (en) | A kind of SAR image target recognition method based on multi-feature fusion | |
| CN104978567A (en) | Vehicle detection method based on scenario classification | |
| CN104834909B (en) | A kind of new image representation method based on Gabor comprehensive characteristics | |
| CN105138987A (en) | Vehicle detection method based on aggregation channel characteristic and motion estimation | |
| CN104134068A (en) | Monitored vehicle characteristic representation and classification method based on sparse coding | |
| CN108205649A (en) | Driver drives to take the state identification method and device of phone | |
| CN106503748A (en) | A kind of based on S SIFT features and the vehicle targets of SVM training aids | |
| CN106778484A (en) | Moving vehicle tracking under traffic scene | |
| CN104008404B (en) | Pedestrian detection method and system based on significant histogram features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181228 |
|
| RJ01 | Rejection of invention patent application after publication |